Google Quantum AI Challenge: Q-gars

AI Disclosure: This Dev post was not written with AI.

Inspiration
- Dicke states are useful in quantum algorithms, and researchers have developed several ways to implement them effectively on both noiseless and NISQ devices (Bärtschi & Eidenbenz, Aktar, Mukherjee). We wanted to understand these methods and adapt them to Google’s Willow chip, accounting for error rates and chip connectivity.
What it does
- QuickeStates is an importable python library which gives users efficient Dicke state preparation circuits tuned to their hardware and error rates. The user specifies their hardware device, desired Dicke state and error rates for CZ gates, single qubit unitaries, and qubit measurements. Across 27 benchmark cases, our circuits used 70.2% fewer CZ gates and 80.7% less circuit depth on average than our naive Bärtschi-SCS baseline while improving fidelity indicators threefold.
How we built it
- We reviewed 5 papers and conducted several experiments to determine which techniques were robust to errors, then handed them off to LLMs for heavy specialized optimization. We explored using Dijkrsta’s algorithm for Bärtschi’s 1D Dicke state implementation, recursive and divide-and-conquer constructions, and delete-tune optimization loops. Several agentic models ran for 15+ hours as we fed them our data and insights from the scientific method.
Challenges we ran into
- One of the biggest challenges we ran into was marginal improvements despite trying a wide range of ideas and lots of compute. Keeping the models working was not sufficient; their progress was impressive but quickly plateaued. This required taking a step back (admittedly, at 3am) and reviewing the core mathematical ideas, which paved the way towards more productive approaches.
Accomplishments that we’re proud of
- We’re proud of producing Willow-specialized Dicke preparation states given arbitrary error rates. We’re also proud of building a agentic workflow to quickly iterate between human insight and numerical optimization. The final result demonstrates successful implementation of cutting-edge circuit preparation techniques on a specific quantum chip.
What’s next
- Next, we plan to implement our Dicke state preparation adapations and optimization loop for other quantum chips and add them to QuickeStates. We’re also looking into automating the process as to optimize for new quantum devices as they become available.